我的数据类似于带有时间戳的多个时间序列。
由group组织，并有一个循环分量，其中time在某些周期内增加，这些周期由这种增加模式的突然变化(即减少)所划分。我想只保留在上升趋势(最后一个上升周期)的最后一次变化之前或之后的数据(行)。
一些合成最小数据:
df <-
data.frame(group = c(rep("A", 10), rep("B", 10), rep("C", 10)),
time = c(c(1:3, 2, 3:6, 5, 6), c(1:2, 1, 3, 7, 6:10), c(4, 3, 6, 4, 6, 7, 6, 8:10))
)
我所说的最后一次上升趋势变化是什么意思:
library(dplyr)
# Just exemplying last change in monotonic increasing trend
df %>%
dplyr::group_by(group) %>%
dplyr::mutate(
row_num = dplyr::row_number(),
time_order = dplyr::case_when(time - dplyr::lag(time, n = 1) >= 0 ~ "increase",
time - dplyr::lag(time, n = 1) < 0 ~ "decrease",
TRUE ~ "increase"),
where_split = dplyr::if_else(dplyr::last(which(time_order == "decrease")) == row_num, "here", NA_character_)
) %>%
print(n = Inf)
#> # A tibble: 30 x 5
#> # Groups: group [3]
#> group time row_num time_order where_split
#> <chr> <dbl> <int> <chr> <chr>
#> 1 A 1 1 increase <NA>
#> 2 A 2 2 increase <NA>
#> 3 A 3 3 increase <NA>
#> 4 A 2 4 decrease <NA>
#> 5 A 3 5 increase <NA>
#> 6 A 4 6 increase <NA>
#> 7 A 5 7 increase <NA>
#> 8 A 6 8 increase <NA>
#> 9 A 5 9 decrease here
#> 10 A 6 10 increase <NA>
#> 11 B 1 1 increase <NA>
#> 12 B 2 2 increase <NA>
#> 13 B 1 3 decrease <NA>
#> 14 B 3 4 increase <NA>
#> 15 B 7 5 increase <NA>
#> 16 B 6 6 decrease here
#> 17 B 7 7 increase <NA>
#> 18 B 8 8 increase <NA>
#> 19 B 9 9 increase <NA>
#> 20 B 10 10 increase <NA>
#> 21 C 4 1 increase <NA>
#> 22 C 3 2 decrease <NA>
#> 23 C 6 3 increase <NA>
#> 24 C 4 4 decrease <NA>
#> 25 C 6 5 increase <NA>
#> 26 C 7 6 increase <NA>
#> 27 C 6 7 decrease here
#> 28 C 8 8 increase <NA>
#> 29 C 9 9 increase <NA>
#> 30 C 10 10 increase <NA>
在2022-05-17由reprex包(v2.0.1)创建
为了便于验证，我提供了我的解决方案:
# All rows until last change in trend, by group
check_until <-
df %>%
dplyr::group_by(group) %>%
dplyr::mutate(
row_num = dplyr::row_number(),
time_order = dplyr::case_when(time - dplyr::lag(time, n = 1) >= 0 ~ "increase",
time - dplyr::lag(time, n = 1) < 0 ~ "decrease",
TRUE ~ "increase")) %>%
dplyr::slice(1:dplyr::last(which(time_order == "decrease"))) %>%
dplyr::select(-c(row_num, time_order))
# All rows after last change in trend, by group
check_after <-
df %>%
group_by(group) %>%
dplyr::mutate(
row_num = dplyr::row_number(),
time_order = dplyr::case_when(time - lag(time, n = 1) >= 0 ~ "increase",
time - lag(time, n = 1) < 0 ~ "decrease",
TRUE ~ "increase")) %>%
dplyr::slice(dplyr::last(which(time_order == "decrease")):max(row_num)) %>%
dplyr::select(-c(row_num, time_order))
我的解决方案有效，但它们似乎太啰嗦和效率低下。我相信还有更优雅的解决方案。欢迎任何见解，我也对datatable解决方案持开放态度。
cumsum+diff+slice(_max)都可以实现。
(1)所有行直到最后的变化趋势:
df %>%
group_by(group) %>%
slice(1:which.max(cumsum(c(1, diff(time) < 0)))) %>%
ungroup()
# # A tibble: 22 × 2
# group time
# <chr> <dbl>
# 1 A 1
# 2 A 2
# 3 A 3
# 4 A 2
# 5 A 3
# 6 A 4
# 7 A 5
# 8 A 6
# 9 A 5
# 10 B 1
# 11 B 2
# 12 B 1
# 13 B 3
# 14 B 7
# 15 B 6
# 16 C 4
# 17 C 3
# 18 C 6
# 19 C 4
# 20 C 6
# 21 C 7
# 22 C 6
(2)趋势变化之后的所有行:
df %>%
group_by(group) %>%
slice_max(cumsum(c(1, diff(time) < 0))) %>%
ungroup()
# A tibble: 11 × 2
# group time
# <chr> <dbl>
# 1 A 5
# 2 A 6
# 3 B 6
# 4 B 7
# 5 B 8
# 6 B 9
# 7 B 10
# 8 C 6
# 9 C 8
# 10 C 9
# 11 C 10